Some Assorted Thoughts On AI

Some thoughts on AI, Large Language Models (LLMs), and Machine Learning, in no particular order.

AI

• From all the discussion around AI, it seems like AI in the workplace is a form of automation. Of course the canonical guide to how automation can disrupt the workplace in the United States is Player Piano.
• There is a very small slice of the workforce of human workers who can be basically replaced one-to-one with machines or software. These people know it, so do their bosses.
• I simply do not believe that AI will become sentient and destroy humanity.
• I saw a question for someone asking about AI use in “3rd and 4th party dependencies” in some software systems. AI might just force companies to understand in details how their software dependencies work.
• AI might just change professional software development the most out of any industry, but that remains to be seen.
• As with many things, artists will likely show us the true value of AI, in some form another.
• I don’t even know what to do or say about OpenAI.
• Like other software products, software quality is a thing that exists and matters. One day I’m sure a lot of Silicon Valley AI companies will realize this.
• AI also seems to be something that will expose the sheer weirdness of humanity, with products like [this](https://www.404media.co/i-tested-the-ai-that-calls-your-elderly-parents-if-you-cant-bothered/).

LLMs

• Note to Self: Learn some machine learning stuff in Python.
• Remember Big Data? Hadoop? These used to be hot technologies. Large Language Models may follow as similar trajectory as folks figure out they don’t necessarily need the “large” part of them.
• LLMs are truly disruptive, and will probably be incorporated into lots of things over time. I think this is largely a good thing.
• Unlike the AI hype, LLMs have been building as an area of research and practice in data science and software engineering for a while now.

Machine Learning

• One of the most memorable demos I’ve seen in a software development/engineering context was for a machine learning algorithm. In the first half of the demo, there was explanation of the algorithm, its assumptions and models used. This part made effectively no sense to me since I’m not really too deep in data science. The second half showed how this model could be directed to a plain spreadsheet of data containing fake employee data, with sheets containing lists of employees by office location, salary, department, and so on. The model took a plain English language query and returned data, and was frankly amazing. It just worked, flawlessly. I feel like this is still how a lot of people who aren’t experts view machine learning today.